[ENH] Adapt the DCNN Networks to use Weight Norm Wrappers

aeon/clustering/deep_learning/_ae_dcnn.py

@@ -296,7 +296,8 @@ def _fit(self, X):
 
         try:
             self.model_ = tf.keras.models.load_model(
-                self.file_path + self.file_name_ + ".keras", compile=False
+                self.file_path + self.file_name_ + ".keras",
+                compile=False,
             )
             if not self.save_best_model:
                 os.remove(self.file_path + self.file_name_ + ".keras")

aeon/networks/_ae_dcnn.py

@@ -241,20 +241,25 @@ def _dcnn_layer(
     ):
         import tensorflow as tf
 
+        from aeon.utils.networks.weight_norm import _WeightNormalization
+
         _add = tf.keras.layers.Conv1D(_num_filters, kernel_size=1)(_inputs)
-        x = tf.keras.layers.Conv1D(
-            _num_filters,
-            kernel_size=_kernel_size,
-            dilation_rate=_dilation_rate,
-            padding=_padding_encoder,
-            kernel_regularizer="l2",
+        x = _WeightNormalization(
+            tf.keras.layers.Conv1D(
+                _num_filters,
+                kernel_size=_kernel_size,
+                dilation_rate=_dilation_rate,
+                padding=_padding_encoder,
+            )
         )(_inputs)
-        x = tf.keras.layers.Conv1D(
-            _num_filters,
-            kernel_size=_kernel_size,
-            dilation_rate=_dilation_rate,
-            padding=_padding_encoder,
-            kernel_regularizer="l2",
+        x = _WeightNormalization(
+            tf.keras.layers.Conv1D(
+                _num_filters,
+                kernel_size=_kernel_size,
+                dilation_rate=_dilation_rate,
+                padding=_padding_encoder,
+                activation=_activation,
+            )
         )(x)
         output = tf.keras.layers.Add()([x, _add])
         output = tf.keras.layers.Activation(_activation)(output)

aeon/networks/_dcnn.py

@@ -146,21 +146,25 @@ def _dcnn_layer(
     ):
         import tensorflow as tf
 
+        from aeon.utils.networks.weight_norm import _WeightNormalization
+
         _add = tf.keras.layers.Conv1D(_n_filters, kernel_size=1)(_inputs)
-        x = tf.keras.layers.Conv1D(
-            _n_filters,
-            kernel_size=_kernel_size,
-            dilation_rate=_dilation_rate,
-            padding=_padding,
-            kernel_regularizer="l2",
+        x = _WeightNormalization(
+            tf.keras.layers.Conv1D(
+                _n_filters,
+                kernel_size=_kernel_size,
+                dilation_rate=_dilation_rate,
+                padding=_padding,
+            )
         )(_inputs)
-        x = tf.keras.layers.Conv1D(
-            _n_filters,
-            kernel_size=_kernel_size,
-            dilation_rate=_dilation_rate,
-            padding="causal",
-            kernel_regularizer="l2",
-            activation=_activation,
+        x = _WeightNormalization(
+            tf.keras.layers.Conv1D(
+                _n_filters,
+                kernel_size=_kernel_size,
+                dilation_rate=_dilation_rate,
+                padding=_padding,
+                activation=_activation,
+            )
         )(x)
         output = tf.keras.layers.Add()([x, _add])
         output = tf.keras.layers.Activation(_activation)(output)

aeon/utils/networks/weight_norm.py

@@ -5,6 +5,7 @@
 if _check_soft_dependencies(["tensorflow"], severity="none"):
     import tensorflow as tf
 
+    @tf.keras.utils.register_keras_serializable(package="aeon")
     class _WeightNormalization(tf.keras.layers.Wrapper):
         """Apply weight normalization to a Keras layer."""